Wall Street giants bet Nvidia’s AI chips will defy the laws of finance

Wall Street is betting that AI chips can defy one of finance’s basic rules: that fast-moving technology quickly loses its value.

Nvidia unveiled this week a $500bn deal under which tech groups will be able to lease semiconductors with financing from groups including Apollo Global, KKR, Brookfield, BlackRock and Goldman Sachs.

The blockbuster partnership was underpinned by expectations that chip prices would remain higher for longer than many analysts had anticipated as a result of the clamour to secure the components powering the AI boom, finance executives involved in the deal told the FT.

Nvidia chief Jensen Huang has said the pact will create a new asset class underpinned by chips, which would be ripe for investment from the $22tn private capital industry.

Investment industry titans, including Blackstone’s Jon Gray and BlackRock’s Larry Fink, have similarly said that their firms are keen to deploy capital to tap into the insatiable appetite for new data centres to train and power the latest AI models.

But the new funding model also comes with the risk that financiers misjudge the durability of demand for — and the value of — Nvidia’s chips.

“The whole thing is predicated on continual investment,” said Ben Bajarin, a technology analyst at Silicon Valley-based Creative Strategies. “There’s a risk of overbuild, that demand eases, that models improve and don’t need as much compute.”

Unlike rental cars or commercial jets, which have a reliable pool of buyers if a customer defaults, chips’ long-term value remains in question as the technology evolves quickly and demand fluctuates.

For debt backed by chip leases, many lenders currently require the loans to be fully paid off within three to five years, under the assumption that the value of the underlying asset will be minimal after that.

Buyers or lessees of chips also have to construct data centres, meaning anticipated revenue streams lag far behind their spending plans.

Huang, however, has argued that even Nvidia’s older H100 chips are holding their value longer than many forecast due to breakneck demand for computing power. In a post on X on Monday evening, Huang said the rental pricing for its recent chips had increased, and even its six-year-old A100 chips were still in use, longer than initial estimates of their lives.

Nvidia’s most powerful chips, which are also known as graphics processing units or GPUs, remain in high demand from AI labs training cutting-edge models, but tech groups are willing to use older hardware for simpler tasks such as answering queries.

Nvidia has also extended the life of chips by regularly updating its software platform Cuda. The platform is central to Nvidia’s dominance in AI processing, enabling customers to use GPUs, originally designed for graphics, to speed up AI applications.

Bajarin said rampant demand for chips of all kinds was likely to bolster Nvidia’s scheme in the medium term and that longer term its success would depend on continuing to lower the total cost of ownership for its products.

Nvidia will also guarantee that the chips hold at least 25 per cent of their value through the lease term, putting the $5.3tn chip giant on the hook for any of their initial losses.

“There is an intrinsic value to the GPU. Nvidia has a decade-long track record on what the rental unit for a GPU is,” one executive participating in the $500bn Nvidia deal told the FT, referring to current rates on A100 chips. They characterised Nvidia’s guarantee as “the first lost piece” of coming financings, meaning the company would absorb early losses in the value of chips beyond certain projections.

Nvidia’s guarantee has led private capital firms to believe that chip financing deals can be packaged into securities and sold to debt buyers such as insurance companies whose assets they manage, tapping large pools of capital inside firms such as Apollo, KKR and Brookfield.

“Private capital has a unique role here,” said another executive partnering with Nvidia, who predicted GPU leases would standardise pricing and create efficiencies for AI groups. Public debt investors would eventually enter the space, making the asset class easier to trade as “people begin to figure it out”, they said.

The structure of the deals will vary between financial companies. Some firms may create speciality finance companies for chips and structure their loans similarly to vehicles used to fund private equity buyouts, known as collateralised loan obligations. CLOs are broken up into multiple rating slices, allowing investors to choose between different risk levels.

Recommended

“You can capitalise the GPUs in different layers of risk, sort of like a CLO,” said one executive. But others balked at that approach, with one noting that “everyone will fund it differently”.

The Wall Street financing could also solve another difficulty for Nvidia, relieving the pressure to finance its customers.

The support from the world’s largest financial institutions would mean that Nvidia could pivot away from vendor financing, through which Nvidia provides direct guarantees to help its clients raise debt in the capital markets, Vivek Arya, a research analyst at BofA Securities, said this week.

Nvidia’s deal with the Wall Street heavyweights was positive for the chipmaker, as “the burden sits with the consortium, not [Nvidia’s] balance sheet”.